Automatically wrangling spreadsheets into machine learning data formats

Abstract

To help automate the important pre-processing step in machine learning and data mining, we introduce synth-a-sizer, a tool for semi-automatically wrangling spreadsheets into attribute-value format, so that they can be used by popular machine learning tools, only requiring the user to mark cells belonging to one single example. synth-a-sizer is based on inductive programming principles. We introduce synth-a-sizer’s transformations, search algorithm as well as a heuristic and distance measure for identifying types. We also report on a first experimental evaluation.